ddx/ddy fails to compile when passed a double literal

Description
Shaders that pass a constant double literal to a pixel derivative instruction fail to compile.

The example uses ddx but it also happens for ddy, ddx_fine, ddy_fine, ddx_coarse, ddy_coarse and fwidth.

Steps to Reproduce
Shader:

[shader("pixel")]
void PSMain() {
    if (ddx(7.5) != 0.0) discard;
}

Command: dxc.exe ddx_constant.hlsl -T ps_6_6 -E PSMain

Actual Behavior

error: validation errors
ddx_constant.hlsl:7:9: error: DXIL intrinsic overload must be valid.
note: at '%1 = call double @dx.op.unary.f64(i32 83, double 7.500000e+00)' in block '#0' of function 'PSMain'.
Validation failed.

I would expect this to work - works fine with FXC. Or at least a better error message.

The program compiles fine if you change the 7.5 to 7.5f.

Environment

• dxcompiler.dll: 1.9(5184-21d28f72)(1.9.2602.17) - 1.9.2602.17 (21d28f727)
• Windows 10

Start by reproducing the issue with ddx_constant.hlsl, entry point PSMain, and dxc.exe ddx_constant.hlsl -T ps_6_6 -E PSMain. Trace how the compiler handles the listed pixel derivative intrinsics when given a double literal versus a float literal. Done means the shader compiles successfully, or reports a clearer diagnostic, for ddx, ddy, the fine and coarse variants, and fwidth.